##
Shifts towards Earlier Onset of Major Depression

In a re-investigation of the NIMH Psychobiology of Depression data, we have studied the question
of shifts towards earlier onset together with the question of steadily increasing lifetime risk
of major depression in successive birth-cohorts. Using a contingency-table approach, it turned
out that inhomogeneities with respect to successive birth-cohorts exclusively showed up in the
neighborhood of principally unobservable combinations of the variables under investigation.
Standard approaches to testing independence in cross-classified data, such as the quasi-independence
model, yielded highly significant results.

## Log-Linear Model

However, through the definition of a log-linear model with weights which replaces the "discrete"
truncation of the quasi-independence approach by a "smoothed" truncation, it was possible to fully
explain the observed age-of-onset shifts, thus supporting the hypothesis that age-of-onset and
birth-cohort are independent.

## Generational Changes in the Lifetime Risk of Depression?

With respect to the question of generational changes in the lifetime risk of depression this
independence implied that such changes should occur at equal rates across all ages of onset. The
analysis yielded significantly larger cohort sizes for the two youngest birth cohorts, a fact which
might be interpreted as an indication of increasing environmental impacts on the genetically
predisposed vulnerability during recent years. However, our cross-sectional survey data were, by
design, not an optimal basis for a reliable assessment of changes in the lifetime risk of depression,
because the risk estimate derived from affected-only survey data corresponds to the probability
that a depressive belongs to a certain birth-cohort and is only loosely related to the lifetime
risk of this cohort (which is the probability that a person belonging to a certain birth-cohort
develops depression).

## Likely the Result of a Statistical Artifact

We therefore conclude, firstly, that method effects are likely to explain a major portion of secular
trends thus far reported in the literature, and, secondly, that there appears to be no clear
necessity to include changing environmental effects into quantitative genetic modelling.